Adaptive Gated Graph Convolutional Network for Explainable Diagnosis of Alzheimer's Disease Using EEG Data

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Electroencephalography (EEG) data analysis is crucial for diagnosing neurological disorders.
• Graph Neural Networks (GNNs) show promise for EEG classification, but their application in Alzheimer's disease (AD) diagnosis is underexplored.
• Existing GNN approaches for AD diagnosis often use basic architectures and functional connectivity for graph inference.

Purpose of the Study:

• To propose a novel Adaptive Gated Graph Convolutional Network (AGGCN) for explainable diagnosis of Alzheimer's disease (AD) using EEG data.
• To develop a GNN model that adaptively learns brain graph structures and enhances node features.
• To improve the accuracy and interpretability of GNN-based AD diagnosis.

Main Methods:

• Developed an Adaptive Gated Graph Convolutional Network (AGGCN) integrating convolution-based node feature enhancement and power spectral density similarity.
• Employed a gated graph convolution mechanism to dynamically adjust contributions from different spatial scales.
• Utilized EEG data for training and evaluating the AGGCN model under both eyes-closed and eyes-open conditions.

Main Results:

• The AGGCN model achieved high diagnostic accuracy for Alzheimer's disease (AD) using EEG data.
• The model demonstrated stable learned representations, performing well in both eyes-closed and eyes-open conditions.
• The AGGCN provided consistent explanations for its predictions, highlighting potential AD-related brain network alterations.

Conclusions:

• The proposed AGGCN offers a powerful and explainable approach for Alzheimer's disease (AD) diagnosis via EEG.
• Adaptive learning of graph structures and dynamic weighting of spatial scales contribute to the model's high performance.
• The explainability feature of AGGCN facilitates further investigation into the neurobiological underpinnings of AD.